[Xenomai-help] Build issue Linux kernel 2.6.38.8 - Xenomai 2.5.6

[Xenomai-help] Build issue Linux kernel 2.6.38.8 - Xenomai 2.5.6

[Jeroen Van den Keybus]

[-- Attachment #1: Type: text/plain, Size: 2045 bytes --]

I have patched a 2.6.38.8 kernel with the Adeos 2.10-01 patch. All Xenomai
2.5.6 built-in in the kernel. The system boots properly on an Atom D525:


> [    0.510257] I-pipe: Domain Xenomai registered.
> [    0.510493] Xenomai: hal/i386 started.
> [    0.510626] Xenomai: scheduling class idle registered.
> [    0.510732] Xenomai: scheduling class rt registered.
> [    0.516816] Xenomai: real-time nucleus v2.5.6 (Wormhole Wizards) loaded.
> [    0.517216] Xenomai: SMI-enabled chipset found, but SMI workaround
> disabled
> [    0.517220]          (check CONFIG_XENO_HW_SMI_WORKAROUND). You may
> encounter
> [    0.517223]          high interrupt latencies!
> [    0.517643] Xenomai: starting native API services.
> [    0.517749] Xenomai: starting POSIX services.
> [    0.517906] Xenomai: starting RTDM services.
>

/proc also contains xenomai entry.

Whenever I run an API function (such as rt_task_create), I get -38 (-ENOSYS
/ Not Implemented) and nothing in dmesg. As such, none of the testsuite
programs can be run.

I have understood from previous posts that compiler options may cause
problems.

What I already tried:

Original:
- Linux kernel compiled for Atom. Stripped down to bare minimums (runs
diskless off NFS).
- scripts/prepare-kernel.sh --linux=../linux-2.6.38.8
--adeos=../adeos-ipipe-2.6.38.8-x86-2.10-01.patch --arch=i686
- ./configure --enable-x86-sep --enable-smp

Variant 1 (no effect):
- ./configure --enable-x86-sep --enable-smp CFLAGS="-m32 -O2" LDFLAGS="-m32"
Variant 2 (reason: kernel seems to be built with -O2 -fomit-frame-pointer
(unsure where to verify though), but no effect):
- ./configure --enable-x86-sep --enable-smp CFLAGS="-m32 -O2
-fomit-frame-pointer" LDFLAGS="-m32"
Variant 3 (no effect)
- Same .config Linux kernel compiled for Pentium Pro (i686)
- ./configure --enable-x86-sep --enable-smp

Since kernel compiles are a very time-costly way of debugging and I also do
not fully understand what I'm really doing when fiddling with gcc options,
it seems that help is now much appreciated.

Thanks,

[-- Attachment #2: Type: text/html, Size: 2575 bytes --]

Re: [Xenomai-help] Build issue Linux kernel 2.6.38.8 - Xenomai 2.5.6

[Gilles Chanteperdrix]

On 09/04/2011 09:05 PM, Jeroen Van den Keybus wrote:
> Variant 2 (reason: kernel seems to be built with -O2 -fomit-frame-pointer
> (unsure where to verify though), but no effect):
> - ./configure --enable-x86-sep --enable-smp CFLAGS="-m32 -O2
> -fomit-frame-pointer" LDFLAGS="-m32"
> 
> Since kernel compiles are a very time-costly way of debugging and I also do
> not fully understand what I'm really doing when fiddling with gcc options,
> it seems that help is now much appreciated.

You need to compile with -fno-omit-frame-pointer, that is precisely the
contrary of -fomit-frame-pointer

-- 
                                                                Gilles.

Re: [Xenomai-help] Build issue Linux kernel 2.6.38.8 - Xenomai 2.5.6

[Jeroen Van den Keybus]

[-- Attachment #1: Type: text/plain, Size: 514 bytes --]

>
>
>
> You need to compile with -fno-omit-frame-pointer, that is precisely the
> contrary of -fomit-frame-pointer
>
>
Indeed I had to, thank you very much. So for sake of completeness:

Variant 4 (works):
- ./configure --enable-x86-sep --enable-smp CFLAGS="-m32 -O2
-fno-omit-frame-pointer" LDFLAGS="-m32"

Still not fully understanding what I'm doing here, but going to look up the
effects pf 'omit-frame-pointer'.

Thanks again,


> --
>                                                                Gilles.
>

[-- Attachment #2: Type: text/html, Size: 1072 bytes --]